Is my IFC file uploaded anywhere?

No. Files are parsed entirely in your browser via web-ifc, a WebAssembly build of the open-source IFC parser. Nothing leaves your machine. The app works completely offline once loaded.

Can I use it with confidential models under NDA?

Yes. Because all processing happens in your browser and no data is ever transmitted to any server, you can safely use it with confidential project models under NDA. There is no service agreement to review with your client — no data changes hands.

Can I share the validation results with my team or client?

Yes. After running validation, click Share Report to copy a link that encodes the full Health Score and issue list directly in the URL — no file upload, no server, no account required. You can also export results as JSON, CSV, or a Validation Certificate. The IFC file itself never leaves your browser.

Yes, fully. IFC Viewer Online works on any modern browser — Chrome, Edge, Firefox, and Safari on both Mac and Windows. No plugin, no installation, no operating system restriction.

What happens if my IFC file is very large?

The app handles files up to 2 GB. Performance depends on your device — files under 200 MB load well on most modern hardware. Once parsed, geometry is cached in OPFS so every subsequent load of the same file is instant, even offline.

How does the validation work?

The validator runs 44 built-in rules in a background Web Worker so the UI stays responsive. Rules cover GUID duplicates, spatial hierarchy, orphan elements, naming conventions, missing type assignments, ISO 19650 property sets, LOD completeness, clash detection, proxy overuse, coordinate offset, storey-elevation integrity, unit consistency, and more. You can also run a buildingSMART IDS specification against the model. Every model receives a Health Score from 0 to 100. Results stream in as each rule completes.

Can I fix errors and re-export a corrected IFC?

Yes. All edits (GUID fixes, renames, property value changes) are held as non-destructive diffs with full undo/redo. Clicking Export IFC applies all diffs to the original binary in a Web Worker and downloads the corrected file — no server involved.

What IFC versions are supported?

IFC2x3, IFC4, IFC4x1, and IFC4x3. The viewer uses the open-source web-ifc parser — the same engine used by most modern BIM tools and @thatopen/components. Outdated schemas are automatically flagged by the validator.

What measurement tools are available?

You can measure lengths between any two points, areas of surfaces, volumes of enclosed elements, and individual edge lengths — all directly in the 3D viewport. Measurements persist as labelled annotations and can be exported alongside your model.

Can I load multiple IFC files at once?

Yes. Load as many IFC files as your device memory allows. Each model has independent visibility, transforms, validation, quantity takeoff, and export. Use the Scene panel to manage, reorder, or remove models at any time.

Does it work offline?

Yes, once the page is loaded. The WebAssembly parser and WebGL renderer run entirely in the browser. Parsed geometry is cached in OPFS so repeat loads are instant even without a network connection.

What are the file size limits?

There is no enforced limit — performance depends on your device. Files up to ~200 MB typically load well on modern hardware. The OPFS geometry cache means you only pay the parse cost once per file.

Is it free? Can I use it for commercial projects?

Yes, completely free. The source code is MIT-licensed. You can use the live app or fork the repository for your own projects, including commercial ones.

Is the inline LiDAR example connected to a physical sensor?

No. It is clearly labelled simulated and deterministic. It validates temporal playback, binary parsing, GPU updates, fault handling and IFC alignment without claiming that a sensor or gateway is connected.

Why keep only two or three point-cloud frames?

For a live view, a fresh frame is more useful than an old one. A bounded newest-valid-frame-wins buffer prevents network jitter or slow rendering from growing latency indefinitely.

Can the custom MCAP recording open as ROS PointCloud2 automatically?

Not automatically. MCAP is a container for timestamped messages with arbitrary serialization. The demo uses an application-owned binary point-frame schema; ROS PointCloud2 and Foxglove PointCloud require explicit adapters.

Back to the BIM & IFC blog

Digital Twins · 2026-08-21 · 12 min

Real-Time LiDAR + IFC in the Browser: MCAP Replay Demo

Run the actual web replay interface and learn the bounded buffers, binary frames and telemetry behind an honestly labelled simulated IFC + LiDAR stream.

Real-Time LiDAR + IFC in the Browser: MCAP Replay Demo — IFC Viewer Online

A video of a point cloud and a live point-cloud stream are not the same product. Video has fixed pixels and mature hardware codecs. A temporal point cloud has coordinates, attributes, poses and timestamps that users may inspect or compare with IFC. That flexibility creates a harder transport and memory problem.

The architecture below treats recordings and live sources as two adapters feeding one bounded frame contract. It lets a team validate playback, seeking, corruption handling and GPU updates with recorded data before purchasing or integrating a sensor.

Five-stage real-time LiDAR web pipeline from sensor and edge filtering through binary transport and a two-to-three-frame buffer to an IFC point-cloud overlay — The browser receives bounded, validated frames; filtering and coordinate normalization belong at the edge.

Try the Temporal IFC + LiDAR Replay

The following live example generates a deterministic scan front around the matching IFC pavilion. The same frame passes through binary encoding, CRC validation and a fixed three-slot transport buffer before it reaches the GPU. Switch to the full viewer to inject deterministic loss, reordering, corruption and a reconnect window.

Operations Pavilion — simulated temporal LiDAR

A 16-second, 12 FPS simulated scan aligned to a bundled IFC4 pavilion. This proves the playback pipeline; it is not a physical-sensor claim.

Actual IFC Viewer Online interface showing a simulated temporal LiDAR replay aligned with the Operations Pavilion IFC — Simulated LiDAR is playing — drag to orbit

IFC Viewer Online displaying a simulated temporal LiDAR scan over the Operations Pavilion IFC with replay and alignment telemetry visible — Actual product capture: the temporal source is a deterministic simulated replay, while parsing, buffering, GPU updates and IFC alignment run in the real viewer.

The Minimum Frame Contract

A transport packet needs more than XYZ values. Sequence numbers reveal loss and reordering. Nanosecond timestamps align points with video, pose and annotations. A declared origin keeps large world coordinates out of Float32 GPU vertices. Attribute flags prevent the UI from pretending that intensity or classification exists when the source never provided it.

Field	Why it exists	Failure caught
sequence	Monotonic frame identity	Loss, duplicate or reorder
timestampNs	Common temporal reference	Video/pose drift
origin + bounds	Local precision and culling	Jitter or absurd allocation
pointCount + stride	Payload budget	Truncated or malicious packet
attribute flags	Explicit semantics	Invented RGB/intensity/class
CRC32	Payload integrity	Corruption in storage or transport

Backpressure: Newest Valid Frame Wins

A reliable network stream can still produce a bad live experience. If decoding takes 120 ms while frames arrive every 80 ms, a normal queue grows forever and the viewer becomes a delayed recording. The correct live policy is bounded: retain two or three reusable slots, reject invalid frames, discard superseded pending work and present latency and drop counts to the user.

Unbounded queue

Latency grows during a slow period
Old frames consume memory and decoder time
The UI still says connected while showing the past
Recovery can take longer than the incident

Bounded live buffer

Memory stays constant
The freshest valid frame replaces stale work
Drops and age remain visible
Recovery is immediate when capacity returns

MCAP for Recording and Reproducible Replay

MCAP is a modular container for timestamped publish/subscribe messages with arbitrary serialization. Its chunks and indexes support lookup by time and topic, which is exactly what a replay timeline needs. The MCAP format specification describes the container and random-access summary records.

Container compatibility is not message compatibility. ROS publishes organized or unordered points through sensor_msgs/PointCloud2; Foxglove defines its own PointCloud schema. The demo MCAP uses a versioned application payload, so importing either ecosystem requires a named adapter rather than hopeful byte casting.

Recorded first

A licensed fixture makes pause, seek, reconnect and regression tests reproducible before hardware integration.

Live adapter second

WebSocket or WebTransport changes delivery, not the frame semantics consumed by the renderer.

Visible telemetry

Latency, frame age, loss, reordering, invalid packets and reconnects belong in the product UI.

Fault injection

Deterministic faults turn conference Wi-Fi failure from a surprise into a rehearsed state transition.

For static measured context, return to the IFC + point-cloud Scan-to-BIM workflow. For a pixel-based resource that can use hardware video codecs and still sit inside the 3D scene, see IFC + video on 3D terrain.

Real-Time LiDAR + IFC in the Browser: MCAP Replay Demo